Stanford researchers have uncovered key organisational factors that determine the success or failure of corporate AI initiatives, following a four-year study that tracked AI implementation at a multinational fashion retailer. The findings reveal how strategic alignment and organizational structure significantly impact ROI on AI investments.
As companies aggressively pursue AI talent—with job postings for machine learning and AI engineers increasing 70-80% in early 2024 compared to 2023—many are discovering that substantial investments aren't translating to successful outcomes. According to Arvind Karunakaran, assistant professor of engineering at Stanford and faculty affiliate at the Stanford Institute for Human-Centered Artificial Intelligence (HAI), "Even as companies invest a lot of money, a lot of projects are failing or not delivering their promised value."
The comprehensive study, conducted between October 2019 and December 2023, followed software developers at a multinational fashion company as they implemented various AI projects. Stanford researchers embedded themselves within the organisation to observe firsthand how developers interacted with business stakeholders and how these interactions ultimately shaped project outcomes.
The study focused on two contrasting AI initiatives that shared the same development team: a successful supply chain distribution optimisation tool and a failed retail productivity optimisation system. By comparing these projects, researchers identified three critical variables that determined success:
Jurisdictional Clarity - The successful project involved a well-defined group of approximately ten allocation specialists who reported to a single manager with clear decision-making authority. Developers knew exactly who to consult and could easily access them. In contrast, the failed project required input from nearly 200 store managers reporting to different finance and district managers, creating confusion about who had relevant expertise.
Task Centrality - For the successful project, product allocation was considered a core responsibility among specialists, motivating them to invest time in developing an AI solution that would directly enhance their daily operations. In the failed case, retail productivity metrics were viewed as peripheral by store managers, who had little incentive to assist with development.
Task Homogeneity - The successful project addressed a standardised process performed consistently across the organisation. The failed project tried to optimise store management practices that varied significantly across locations with unique consumer demographics and employee dynamics, making a uniform AI solution impractical.